2019 Chapter 4:
The Ragnaroks That Weren’t
Yeah, I know, it sounds like a title of a low-rent Doctor Who/Norse Mythology mashup special, but the truth is that it fits.
It fits because I finished one complete Ragnarok 2 design—to cosmetic screened metal—before one tiny fault made me change course, AND I finished a second Ragnarok 2 design—in 3D CAD and in layout—before nagging doubts made me throw the whole thing away and start with a clean-sheet design.
It’s that clean-sheet design that is the new Ragnarok 2 we just previewed at the NYC Can-Jam, and it’s that clean-sheet design that is the one we’ll be shipping soon. This chapter isn’t about that clean-sheet design.
This chapter is an exploration of failure.
This chapter is a warning about playing it too safe.
This chapter is an examination of what could have been, but won’t be.
(And yes, we’ll have pictures.)
Ragnarok 2, Version 0: The Concept
As soon as we launched Ragnarok, we knew there would be a Ragnarok 2.
Surprised? Don’t be. Ragnarok itself was a huge, huge undertaking. It was the most complex product we’d done up to that time. It introduced a whole lot of new ideas, including complete microprocessor management of operating points. It had 5 PC boards (later simplified to 3), lots of internal wiring, a complex chassis, and firmware.
Think of the first Ragnarok as our moonshot. And when you’re working hard on something that’s so different, so much more complex, so much more advanced than what you’ve been doing before, you’re worried most about getting the basics right…you know, you’re worrying about getting your guys to the moon and back alive, not about how many flavors of syrup they’ll have for their space pancakes. With Ragnarok, we were worrying about making the algorithms that run the amp as stable as possible, and the amp itself as robust as possible.
Which meant we missed things. Things like:
- A friggin remote control—this was a speaker amp!
- The ability to switch between headphone output, speaker output, or neither.
- Complexity in assembly issues (which is why Ragnarok got revised for less boards).
The first issue—the lack of remote—was the main thing, but it was an issue simply too big to address. We had never done a remote of any time. At the time we did it, I thought we would want to do a RF or Bluetooth remote, which would increase complexity. Even if we did IR, we didn’t have a good answer for how the remote would interact with the volume knob, and we hadn’t figured out a safe way to make that work. We later came up with a scheme for Saga and Freya, but when you’re deep into a super-complex product, adding something like a remote control is simply too much to add. We knew, though, that someday, somehow, Ragnarok would need a remote.
The second issue—the ability to switch between output modes—we addressed, in a kludgy way, in firmware. Hold down a button and cycle through the output modes. The problem was, there was no indication of which mode you were in. But, by the time we thought to address this, the die was cast, the metal was done, and there wasn’t going to be a better solution without a complete re-work of the front panel. Which would then require a re-screen of the main chassis. Which would be very pricey. Which would probably not be a great thing to offer, unless it came with a remote.
The third issue—complex assembly—we addressed over time. First by creating an entirely separate production team to work on more complicated products like Ragnarok and Yggdrasil. This was actually a good thing, because it got us ready for products that weren’t just a “put a board in a box” level. Second by revising the Ragnarok to integrate three PC boards into one, and reduce the total number of boards from 5 to 3. This “Ragnarok 1.5” ended up being the vast majority of Ragnaroks we made over its total run, but it still didn’t address many of Ragnarok’s complexities: lots of wiring, ribbon cables (more on those later), two transformers, a separate summing stage for single-ended output, standoffs and cages and fiddly assembly. In retrospect, it’s fortunate that Ragnarok didn’t sell too well.
And, of course, we got feedback from people who owned Ragnarok, so we could add some additional issues to the list:
- They didn’t like how big it was
- Or how hot it ran
- Or that the single-ended output was weaker/different than balanced
The first issue—size—makes a bit more sense when you think of Ragnarok as a rack product, not a desk product. I was frankly surprised how many people were using Ragnaroks on desks. That’s a bit nuts. And Ragnarok really needed its size for the big transformer and to dissipate heat. So a small Ragnarok was never in the cards. I thought briefly about doing a “Ragnarok Mini” with Class D…and then listened to some available Class D implementations…and then realized the problem with Class D is really output filtering and how it responds to different speaker impedances…and then realized this was a much bigger project and put it aside.
The second issue, heat, I gotcha. Ragnarok ran hot mainly because MOSFET outputs are most linear when run hard. Unlike bipolars, which have an optimal bias point if you’re shooting for a classic Class AB output stage, MOSFETs are better the hotter they run. Plus, Ragnarok rejected a ton of heat off of the voltage gain stage due to its highly stacked supplies, necessary to provide linear performance for its non-overall-feedback design. The end result was a very hot-running amp. This is something we could address with Ragnarok 2.
The third issue—the use of summers to get single-ended output—again, was unavoidable in Ragnarok. Ragnarok’s circlotron-style output stage is a neat way to get inherently balanced output, but that was also the catch—you couldn’t really use it any other way. Single-ended needed to be created by summing the positive and negative phases of Ragnarok’s circlotron stage. The result? Lower power output into headphones from the SE side, and a slightly different sonic signature. Not ideal. That was something that always bugged me, and I made it a point to address it in Ragnarok 2.
And, of course, over the years of making Ragnaroks, we learned a few things ourselves. Most particularly, it became clear that the microprocessor oversight of operational points was both a blessing and a curse. A blessing because it could keep things tightly regulated, a curse because the amplifier was completely and 100% totally reliant on the microprocessor to tweak bias three times a second, every second, continuously, forever.
Aside: It had to do this because Ragnarok was never designed to be highly isothermal—a fancy engineering-speak word for keeping things at the same temperature. Most power amps go to great lengths to keep the bias circuit isothermal to the outputs (as we do in Vidar—the bias transistors are clamped directly to the same heatsink as the outputs), but Ragnarok, with its microprocessor oversight, didn’t need this. Furthermore, it would have been very, very hard to reengineer it, given the limitations of its circlotron topology and non-overall-feedback design. But if it could be made isothermal, the microprocessor’s job could become a lot easier. Which would make testing Ragnarok easier, because the microprocessor wouldn’t de-bias the output when presented with steady-state sine waves.
So yeah, with those kinds of issues, it seemed pretty clear what Ragnarok 2 should be: a remote-controlled, more verbose, cooler-running, more-isothermal, balanced-and-SE agnostic, simpler/easier-to-assemble version of Ragnarok.
Because, for the gripes, most people liked the way Ragnarok sounded. Oh yeah, some of them said that it wasn’t super-quiet with super-sensitive custom IEMs, but hey, I mean, it was like a 25-pound deskbreaker that fed full speaker output to the balanced headphone jack. You can only do so much there. And some wanted more power out of Ragnarok 2, but the reality was that it was at the limits of a 60WPC amp. Heck, I thought of de-rating it to 50W. Because even if it was cooler-running at idle, it still needed to make it through real-world use…
But I’m getting ahead of myself. Let’s talk about what we did for the first Ragnarok 2 prototype–the one we took all the way through to finished metal.
Yes, all the way to finished metal. Yes, this was expensive. Yes, this was stupid. Yes, this was wasteful.
But hey, I never said we were the most brilliant minds on the planet.
Ragnarok 2, Version 1: The Surprise
Surprise? Yes. But let me get to that.
When I started with the Ragnarok 2 design, I thought I had a pretty clear notion of what it would be. In short, exactly what we outlined above:
- It would have a remote that controlled volume, input select, and output mode.
- It would indicate what output mode you were in, and it would have an extra front panel button to select it.
- It would use a single transformer, only two boards, and would eliminate as much wiring as possible to make it easier to assemble.
- It would run cooler, due to the lower bias requirements of BJTs and lower stack voltage.
- It would use a single gain stage for both SE and balanced outputs—Jotunheim’s Pivot Point topology, to be exact.
- It would also be more isothermal, so we could dramatically reduce the microprocessor oversight of bias.
Well, almost exactly what we outlined. I had one embellishment, and one compromise:
The embellishment: a large optional input card that could turn Ragnarok 2 into a digital AIO, with USB, optical, and coax inputs. This card was physically much larger than the cards that went into Jotunheim, and would be specific to Ragnarok 2.
The compromise: in order to ensure that Ragnarok 2 would not thermally shut off when run into difficult loads, I added a panic fan.
“Oh holy hell a fan, a fan, no no no no, not a fan!” you might be yelling.
And I get it. That’s pretty much my reaction when faced with a fan on a piece of audio gear. You might feel a little more copacetic knowing that it was my intention that the fan was always off. Or at least off for the 99.5% of the time you weren’t blasting 2 Live Crew at 115db into 4 ohm speakers. Or you might not. Because fans, even fans that are intended to stay off, are perhaps a sore point in a product with a 5-year warranty.
But I also wanted to make 100% sure that Ragnarok 2 wouldn’t melt down at any time, and that meant that, when run hard, the passive heatsinking from the aluminum chassis might need a little forced air running past it to keep it at an acceptable temperature.
So, yeah, a fan.
“Wait a sec, hold on the fan,” others might be saying. “What’s this about a bigger optional DAC card? I saw the Ragnarok 2 photos and they have smaller card slots, like Jotunheim and Lyr. I want my BIG MYTHICAL OPTIONAL SUPER DUPER DIGI CARD NOW DAMNIT!”
Yeah, I hear you. And that’s why you should go back and read the chapter about how you can't always get what you want. Bottom line, we can’t have infinite variations.
And, realistically, testing and debugging three major new products (because of course you would want a phono card version too) is massively more complicated than one, and, when you add in the fact that the first thing people asked when confronted with the card choice was to ask if they could have both, and when you add in the fact that we have several perfectly good cards in production, and when you add in the fact that if you want a more capable DAC, we got tons of those too, welllllllll…you begin to see why the super cards died.
Okay. Back to the first Ragnarok 2 design.
Electronically, I knew what I wanted to do. Or so I thought.
- I figured I could use the Pivot Point topology from Jotunheim for the basic gain stage. However, we’d never deployed a Pivot Point stage that could drive speakers, and when you’re talking squirrely, high-bandwidth current-feedback topologies, I figured it could get really interesting. But, Pivot Point would eliminate the need for the summers. If it worked stably, that would be a huge win. Still, I expected it would be a helluva fight to get it to work.
- I also thought it was time to move to the bipolar outputs I was using in Vidar, and to arrange them with SMD drivers mounted on the back of the board with a gap pad like we use in Lyr 3 in a highly isothermal configuration. Those of you who follow along all the chapters are beginning to see some design reuse.
- I really wanted to use a single transformer, preferably on the same core as Vidar; this worked out very, very well; the transformer used in Aegir and Ragnarok 2 are the same except for the potting and shielding—only Ragnarok 2 needs the shield. But I'm getting ahead of myself.
- I wanted to integrate the two front boards to a single board, to reduce the total number of boards to 2, but it turned out that wasn’t going to be; I considered it acceptable, since we’d been making Ragnaroks like that for a while and it wasn’t a huge issue. Or so I thought.
- I also figured it was time to bump up the resistor ladder volume control from 64 to 128 steps.
Mechanically, I also knew what I wanted to do:
- Something similar in appearance to the older Ragnarok, but simpler
- A new front panel board with the extra lights and holes and buttons for remote control
- A chassis that dispensed with the top sled design and went to a U-channel interior piece, to simplify and save a little money
- A nice big hole on the side for The Card That Will Never Be
- A nice big hole on the bottom for the fan (yes, the fan would have sucked air from underneath the Ragnarok)
In short, not too shocking of a change.
But, with the changeover to the new Pivot Point topology, I figured I’d be in for a fight. Low-powered amps get away with a lot of things that high-powered amps don’t. Something that works for 5W into 32 ohms wouldn’t necessarily work for 60W into 8 ohms—you’re talking massively different output devices, biasing scheme, a serious driver stage…all the stuff you need to run way more current into far lower output impedances. I expected to see massive oscillation when I first fired it up, or when I connected it to real speaker loads. With such a fight coming, maybe it’s not surprising that I wanted to play it safe.
Also, I knew that the creation of a new, large digital input card with multiple inputs meant the development of an entirely new product itself—new firmware for both the card’s microprocessor and DSP would be required. Plus, we’d either have to have a very smart auto-switching protocol between all the inputs, or we’d need to figure out a way to indicate which digital input was selected on the front panel. So there was a whole ‘nother wrinkle. More reason to keep the changes minimal.
So I went about doing the first prototype. To start, I laid out a board and drew up a bottom chassis in CAD (because I could bolt it to the current top chassis for thermal testing.) The board layout was fairly easy, and actually allowed me to spread the output devices out farther on the bottom for better thermal dissipation.
Aside: going to Pivot Point meant we were going to be using 2X the output devices of the old Ragnarok—insanely overbuilt for the power output, but hey, I’d rather something be overbuilt than the opposite.
The new PCB layout also let me put the drivers, bias, and voltage gain right in-between each pair of output devices, on the bottom of the board, and couple everything tightly together with a Gap Pad. This proved to be so good that no oversight of bias was necessary at all—set it once and forget it. Hell, we didn’t even need the microprocessor to do that! That was a huge win. The microprocessor could be used more as oversight, rather than as a micromanager.
First fire-up was shockingly uneventful. No oscillation. No smoke. No heat. I biased everything up and the bias just sat there, rock-solid.
It can’t be that easy, I remember thinking. Of course, without firmware, I couldn’t test the amp to any real degree, so I handed it off to Dave for that side of things. I fully expected him to come back in a week or two, shaking his head over a carbonized board, and telling me how it blew up three times on him, and now the board wasn’t fixable anymore.
(You know, like the original Ragnarok.)
Nope. In a week, he brought it back. The board looked unscathed, and Dave seemed happy. Since he wasn’t immediately forthcoming, I asked, kinda cautiously:
“So…..it’s done?”
Dave nodded. “Yep.”
“And…..it works?”
“Seems to.”
Huh. I hadn’t expected that. There was no way we could just start the stupid thing up and expect it to work, could we?
But it did. I plugged it in, ran some tests, and it seemed to work just fine. No oscillation, no smoke, no flames. It even played fine into speakers. I even took it to the Schiitr and had a couple of people listen to it.
I mean, it was, pretty much, well…done.
“So why not just ship it?” you cry. “Hell, I coulda had a Ragnarok 2, like, a year ago!”
Yeah, I hear you. Except:
- When something’s that easy, you’re going to pay for it.
- I was already having indigestion about how safe we’d played it.
And, let’s face it. We were in no way done. Ragnarok 2 Version 1 might have worked, but the remote? No, I needed to design it and order it. The giant DAC card? Not so much. Heck, at that point it didn’t even exist. That was Dave’s fight, and it took a while to get it even in testable form. I’m not sure we ever worked it out 100%. And when we told a few early listeners about the DAC card, they immediately got excited…
…and ask about a phono preamp…
…and if they could have both DAC and phono pre.
ARGH.
Plus, I was already having some indigestion of my own, both in terms of the choice of topology, and in how safe we’d played it.
“Topology? What’s wrong with Pivot Point?” you ask.
In short, not much. Pivot Point is a very nice topology. It does, however, exhibit the N/N+1 gain relationship typical of a multi-stage amp driven from both the positive and negative terminals (though without the low impedance problem on the negative side.)
I always really wanted to have it all: an N/N gain structure, the ability to apply feedback, and high-impedance input terminals. And I had been testing a lot of candidate topologies, trying to find one that worked in the real world, rather than simply in simulation. And, at that point in Ragnarok 2’s development, I was tantalizingly close on a breadboard/hackjob platform on a topology that would become Nexus—which works in the real world with the single caveat of matched devices throughout the stage (not a trivial caveat, but not insurmountable). If I could put that in Ragnarok 2, it would be a huge win. Of course, it also could possibly have the same oscillation/fire/smoke problems I’d worried about with Pivot Point.
Aside: sorry for the tech diversion. A discussion of Pivot Point vs. Nexus vs Mjolnir’s single-gain-stage Circlotron will be the subject of a future chapter.
But this indigestion wasn’t enough to keep me from proceeding. I finished the CAD for Ragnarok 2 and sent out for finished first articles, figuring I could introduce the product fairly quickly—maybe at the LA high-end show, or RMAF. I addressed the topology problem by laying out a next-gen Ragnarok 2 board with The Thing That Would Become Nexus, and did a second prototype.
This one kinda-sorta worked. It had compensation problems, and the DC measurement scheme kinda sucked (which Dave fixed), and it needed a myriad of tweaks, but it worked…and it helped me uncover some of the last problems with Nexus—including the primary need for matched devices.
By this time, the Giant DAC Board was limping, and I’d laid out a Giant Phono Board to go along with it (which was not doing so well, we had huge field problems). We were getting pretty close to something we could actually show.
But…the downsides of playing it safe began coming to the fore. Building a Ragnarok 2, while better than the old Ragnarok, was only slightly better. It was still a crazy nest of wires and cables. It still had three boards, two of which mounted to the front of the chassis with nerve-wracking blind tapped holes. And it still had the complicated, pain-in-the-rear-end, expensive chassis. I liked it less and less as time went on.
The final straw was the ribbon cables.
Yes. Something that simple. A standard ribbon cable we buy by the thousands. Dead-simple, old-tech, should be 100% reliable, right?
Well, no.
In fact, I spent two days trying to debug something with Ragnarok 2 that wasn’t even a problem. It was just a crappy ribbon cable.
And, when I realized that, I sat back in my chair and looked at the mess of wires and ribbons and boards, and thought,
This is stupid! I almost didn’t fix anything here!
I mean, if I could use a single board, rather than the stupid 2-board arrangement up front, maybe we could get rid of all the extra headphone wiring. Hell, even better, if it was a horizontal board, we might be able to use pin headers and eliminate the friggin ribbon cables entirely. Hell, it might be able to be held together with just pin locators and a single nut on a motorized pot, so Ragnarok’s volume setting could track the remote…
“Oh hell,” I said, and put my head in my hands.
Because all of this was totally doable.
And if I hadn’t been so scared of the design, I would have simply done it. And it would be way simpler. And it would be right.
Crap.
I can still do this, I thought. All it would mean is throwing away some metal. And a whole lot of very precise calculations to make sure everything lined up. And it would look a bit different than the old Ragnarok. But that was OK, because this was the right thing to do.
So I went back to the drawing board…
Ragnarok 2, Version 2: The Stillborn
Before I go into this, you need to know how big a deal it was, in terms of our mechanical engineering capability. Because, before this, the most challenging thing we attempted was the slide-lock fasteners on the top of Jotunheim and Lyr 3. What I was thinking about doing was a complex mechanical sub-assembly that had to precisely align to the Ragnarok 2 motherboard, mate with a stiffening subchassis, be assembled with light-pipes and connectors in situ, locate into the front panel with tangs, and be held together with a single nut (on the potentiometer).
Whew. Put like that, it’s no wonder I started with simulations, rather than going straight to 3D CAD.
No, instead I started in 2D, laying out super-detailed models of all the parts on the board, on the riser, on the subchassis, with the standoffs, boards, etc all in place—to see if it would fit, how it would interact with the main board, what would happen if things were a bit off.
To make a long story short, it fit. It even allowed us to use the motorized pot I wanted to use, so volume could finally track the remote.
I went to 3D CAD, finished up the chassis changes, designed the subchassis, and sent everything out for quote. At the same time, I laid out the new horizontal mezzanine board, and sent it out for prototype.
When the quotes for the chassis came back, I was working on the main board, adding the pin headers (which turned out to be a standard part we were already using—big win) and getting ready to send it out for prototype.
I looked at the chassis quotes. They were as I expected, pretty much the same as Ragnarok 1 and Ragnarok 2.
Ah well, can’t be helped, I thought.
It’s an expensive chassis, nothing we could do about it.
After I finished the board, I dropped a 2D version of it down on the 2D fit-test drawing I’d been using to make sure everything lined up. It sat perfectly under the mezzanine board pins, and didn’t get anywhere near the fan. We were good.
The fan…
On top of an expensive chassis, I’d have to find a quiet fan. Because if it ever came on, it would have to be at least acceptable. And it would have to be reliable with intermittent use, so when it was supposed to come on, it would actually do so, even if it had been sitting for 5 years.
So yeah, a few more bucks for a fan…
Yeah, not the end of the world, but people hated fans, and I hated fans, and nobody in their right mind liked fans, even if they were usually off.
But there was nothing we could do, I thought.
It wasn’t like we could use passive heatsinks like Vidar.
Even though the heatsinks were about the same size...
Even though the heatsinks would dramatically simplify the chassis...
Even though they might literally be exactly the same heatsinks we were already using in Vidar...
Even though that kind of design re-use might make the chassis less costly.
I rocked back in my chair again, thunderstruck. Holy friggin hell, if we could use Vidar heatsinks, that would literally solve every single problem with the old Ragnarok…including the costly, fiddly, hard-to-assemble chassis. And it would increase the number of heatsinks we were already using.
But…that was a
complete redesign.
And…I was, pretty much,
done.
In that moment, I could literally see the demon and the angel sitting on my shoulders, like the silly old cartoons. The demon was laughing his ass off, saying,
**** it, you’re done, go have a beer, it’s better than it was, whadya want to work the rest of your life? And the angel was there, just looking at me and saying nothing, because it didn’t have to, just a look was enough.
I needed to do the right thing.
And the right thing was tearing everything up, throwing it away, and starting with a clean sheet. So that Ragnarok 2 could be really, really, really right.
But that’s a story for another chapter.
, variac?) so I don't know how they are routed, but I'm sure Jason would be doing a better job than I could imagine PSU noise-wise.
